Fault diagnosis of PEMFC based on the AC voltage response and 1D convolutional neural network

Abstract

Real-time diagnosis is required to ensure the safety, reliability, and durability of the polymer electrolyte membrane fuel cell (PEMFC) system. Two categories of methods are (1) intrusive, time consuming, or require alterations to the cell architecture but provide detailed information about the system or (2) rapid and benign but low-information-yielding. A strategy based on alternating current (AC) voltage response and one-dimensional (1D) convolutional neural network (CNN) is proposed as a methodology for detailed and rapid fuel cell diagnosis. AC voltage response signals contain within them the convoluted information that is also available via electrochemical impedance spectroscopy (EIS), such as capacitive, inductive, and diffusion processes, and direct use of time-domain signals can avoid time-frequency conversion. It also overcomes the disadvantage that EIS can only be measured under steady-state conditions. The utilization of multi-frequency excitation can make the proposed approach an ideal real-time diagnostic/characterization tool for fuel cells and other electrochemical power systems.